GigaSpeech 3 for Arabic speech recognition
Original title: GigaSpeech 3: Scaling Multi-Dialect Arabic Speech Recognition with Weakly Labeled Data
39,000-hour GigaSpeech 3 corpus covers five Arabic dialects and reduces WER by 16.6% relative to prior systems.
GigaSpeech 3 is a 39,000-hour corpus with 10 million weakly transcribed segments for Algerian, Egyptian, Iraqi, Saudi, and Syrian Arabic. Dialect-specific models achieve 23.86% macro-average WER. A multi-dialect model using self-predicted dialect identities reaches 25.51% WER.
Unconfirmed: Corpus release and evaluation metrics as stated by authors.
- 39,000 hours of speech data
- 10 million weakly transcribed segments
- Macro-average WER of 23.86% for dialect-specific models
- Outperforms third-party systems on three dialects
- Supports Modern Standard Arabic in multi-dialect setup
- Organizations
- Hugging Face
- AI models
- GigaSpeech 3
From the original article
1 min read at huggingface.co
Dialectal Arabic speech recognition remains constrained by the scarcity of dialect-resolved transcribed speech despite the abundance of web audio. We introduce GigaSpeech 3, a corpus of 39,000 hours and 10 million weakly transcribed speech segments covering Algerian, Egyptian, Iraqi, Saudi, and Syrian Arabic.
By Junzhe Liu, Yujie Tu, Yifan Yang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.